Microprocessor-controlled integrated circuits are used in a wide variety of applications. Such applications include, for example, personal computers, vehicle control systems, telephone networks, and a host of consumer products. As is well known, microprocessors are essentially generic devices that perform specific functions under the control of a software program. This program is stored in one or more memory devices that are coupled to the microprocessor. Not only does the microprocessor access memory devices to retrieve the program instructions, but it also stores and retrieves data created during execution of the program in one or more memory devices.
There are a variety of different memory devices available for use in microprocessor-based systems. The type of memory device chosen for a specific function within a microprocessor-based system depends largely upon what features of the memory are best suited to perform the particular function. For instance, random access memories such as dynamic random access memories (DRAMs) and static random access memories (SRAMs) are used to temporarily store program information and data “actively” being used by the microprocessor. The data stored in random access memories may be read, erased, and rewritten many times during the execution of a program or function. Read only memories (ROMs), “write once read many” devices (WORMs) and electrically erasable programmable read only memories (EEPROMs), and flash memories on the other hand, are used as long term memory devices which permanently store information about the microprocesor system or store software programs or instructions for performing specific functions until erased or deleted by a user, for example.
Random access memories tend to provide greater storage capability and programming options and cycles than read only memories, but they must be continually powered in order to retain their content. Most random access memories store data in the form of charged and discharged capacitors contained in an array of memory cells. Such memory cells, however, are volatile in that the stored charges will dissipate after a relatively short period of time because of the natural tendency of an electrical charge to distribute itself into a lower energy state. For this reason, most random access memories such as DRAMs must be refreshed, that is, the stored value must be rewritten to the cells, about every 100 milliseconds in order to retain the stored data in the memory cells. Even SRAMs, which do not require refreshing, will only retain stored data as long as power is supplied to the memory device. When the power supply to the memory device is turned off, the data is lost.
Read only memories presently used in microprocessor devices are non-volatile, that is, capable of retaining stored information even when power to the memory device is turned off. Some read only memory devices are constructed so that once programmed with data, they cannot be reprogrammed. The read only memories that can be reprogrammed have complex structures which are difficult to manufacture, occupy a large amount of space and consume large quantities of power. For these reasons, read only memories are unsuitable for use in portable devices and/or as substitutes for the frequently accessed random access memories, i.e., memories capable of 1014 programming cycles or more.
Efforts have been underway to create a commercially viable memory device that is both randomly accessed and nonvolatile. To this end, various implementations of such nonvolatile random access memory devices are presently being developed which store data in a plurality of memory cells by structurally or chemically changing the resistance across the memory cells in response to predetermined voltages respectively applied to the memory cells. Examples of such variable resistance memory devices include memories using variable resistance polymers, perovskite, doped amorphous silicon or doped chalcogenide glass.
In a variable resistance memory cell, a first value may be written thereto by applying a voltage having a predetermined level to the memory cell, which changes the electrical resistance through the memory cell relative to the condition of the memory cell prior to the application of the voltage. A second value; or the default value, may be written to or restored in the memory cell by applying a second voltage to the memory cell, to thereby change the resistance through the memory cell back to the original level. The second voltage is in the negative direction of the first voltage and may or may not have the same magnitude as the first voltage. Each resistance state is stable, so that the memory cells are capable of retaining their stored values without being frequently refreshed. In this regard, since the variable resistance materials can be “programmed” to any of the stable resistance values, such variable resistance memory cells are known as programmable conductor random access memory (PCRAM) cells.
The value of the PCRAM cell is read or “accessed” by applying a read voltage to determine the resistance level across the cell. The magnitude of the read voltage is lower than the magnitude of the voltage required to change the resistance of the PCRAM cell. In a binary PCRAM cell, upon determining the resistance level of the PCRAM cell, the detected resistance level is compared with a reference resistance level. Generally, if the detected resistance level is greater than the reference level, the memory cell is determined to be in the “off” state, or storing, for example, a value of “0.” On the other hand, if the detected resistance level is less than the reference level, the memory cell is determined to be in the “on” state, or storing, for example, a value of “1.”
FIG. 1 generally shows a basic composition of a PCRAM cell 10 constructed over a substrate 12, having a variable resistance material 16 formed between two electrodes 14, 18. One type of variable resistance material may be amorphous silicon doped with V, Co, Ni, Pd, Fe and Mn as disclosed in U.S. Pat. No. 5,541,869 to Rose et al. Another type of variable resistance material may include perovskite materials such as Pr1-xCaxMnO3 (PCMO), La1-xCaxMnO3 (LCMO), LaSrMnO3 (LSMO), GdBaCoxOy (GBCO) as disclosed in U.S. Pat. No. 6,473,332 to Ignatiev et al. Still another type of variable resistance material may be a doped chalcogenide glass of the formula AxBy, where “B” is selected from among S, Se and Te and mixtures thereof, and where “A” includes at least one element from Group IIIA (B, Al, Ga, In, Tl), Group IVA (C, Si, Ge, Sn, Pb), Group VA (N, P, As, Sb, Bi), or Group VIIA (F, Cl, Br, I, At) of the periodic table, and with the dopant being selected from among the noble metals and transition metals, including Ag, Au, Pt, Cu, Cd, Ir, Ru, Co, Cr, Mn or Ni, as disclosed in U.S. Published Application Nos. 2003/0045054 and 2003/0047765 to Campbell et al. and Campbell, respectively. Yet another type of variable resistance material includes a carbon-polymer film comprising carbon black particulates or graphite, for example, mixed into a plastic polymer, such as that disclosed in U.S. Pat. No. 6,072,716 to Jacobson et al. The material used to form the electrodes 14, 18 can be selected from a variety of conductive materials, such as tungsten, nickel, tantalum, titanium, titanium nitride, aluminum, platinum, or silver, among others.
In FIG. 2, a PCRAM cell 100 is shown to include an access device 102, a programmable conductor memory element 104, and a cell plate 110. The access device 102 is a transistor having a gate 102a coupled to a word line 106 and one terminal (source) 102b coupled to a bit line 108. The other terminal (drain) 102c of the access device 102 is coupled to one end of the programmable conductor memory element 104, while the other end of the programmable conductor memory element 104 is coupled to the cell plate 110. The cell plate 110 may span and be coupled to several other PCRAM cells, and may form the anode of all the memory elements 104 in an array of PCRAM cells. The cell plate 110 is also coupled to a potential source 112.
A representative diagram of the physical structure of the memory cell 100 is shown in FIG. 3. In particular, an n-doped substrate 126 includes a p-doped region 124. Two n-wells 120, 122 are formed in the p-doped region 124. Access device 102 is formed on the surface of the substrate 126 between the two n-wells 120, 122, so that the two n-wells 120, 122 serve as the source 102b and drain 102c respectively, of the access device 102. Word line 106 is formed as a conductive strip extending into the page across the top of access device 102. Bit line 108 is connected directly to the n-well 120 forming the source 102b of the access device-102. Programmable conductor memory element 104 is formed on the substrate with its cathode 114 in contact with n-well 122 and the cell plate 110 (only a portion of which is shown) as its anode. The cell plate 110 of memory element 104 is connected to a potential source 112.
In the conventional operating scheme, when the memory element 104 is idle, the voltage across the anode 110 and the cathode 114 is 0 V. In order to perform any access operations including programming the programmable conductor memory element 104 to the low resistance state, erasing a programmed memory element (i.e., returning the memory element to the high resistance state), or reading the value stored in memory element 104, the word line 106 must be activated by applying a threshold voltage VG. The activated word line which closes the gate 102a of the access device 102 so that an n-channel is formed in the substrate 126 under the gate structure of access device 102 and across the gap between the two n-wells 120, 122 to activate the device 102. Upon activating the access device 102, the memory element 104 can be programmed to the low resistance state by applying a write (positive) voltage having at least the magnitude of a threshold voltage VT across the memory element 104.
Application of the write voltage may be achieved by raising the potential at the cell plate 110 (anode) relative to the access device drain 102b by applying or raising the voltage at the potential source 112, lowering the potential of the bit line 108, or a combination of both. To erase a programmed memory element 104, a negative voltage having a magnitude of at least a threshold erase voltage is applied between the anode and the cathode of the memory element 104, such that the potential at the potential source 112 is lower than the potential of the bit line 108.
The value stored in the memory element 104 can be read by applying a positive voltage, either by raising the potential at the anode or lowering the potential of the bit line 108 or both, which is less than the voltage level required to program the memory element to the low resistance state.
In a bistable PCRAM cell, the programmable conductor memory element 104 stores, for example, a binary 0 when in a high resistance state, and a binary 1 when in the low resistance state. The memory element 104 is ideally programmed to the low resistance state, e.g., to store a binary 1, by applying a positive voltage having a magnitude at least that of a threshold write voltage, and can be restored to the high resistance state, e.g., to store a binary 0, by applying a negative voltage having a magnitude of at least a threshold erase voltage. Of course, the values “0” and “1” relative to the high and low resistance state, respectively, are user-defined, and thus could be reversed, with the high resistance state representing the value “1” and the low resistance state representing the value “0.” The memory element can be nondestructively read by applying a reading voltage having a magnitude of less than the threshold write voltage.
As with volatile RAMs, PCRAMs are arranged as an array of memory cells and are written, erased, and read using a controller. FIG. 4 illustrates a memory device 200 comprising an array of memory cells arranged by tiling a plurality of memory cells 100a-100f together so that the memory cells along any given bit line 108a, 108b do not share a common word line 106a-106c. Conversely, the memory cells 100a-100f along any word line 106a-106c do not share a common bit line 108a-108b. In this manner, each memory cell is uniquely identified by the combined selection of the word line to which the gate of the memory cell access device is connected, and the bit line to which the drain of the memory cell access device is connected.
Each word line 106a-106c is connected to a word line driver 202a-202c via a respective transistor 204a-204c for selecting the respective word line for an access operation. The gates of the transistors 204a-204c are used to selectively couple/decouple the word lines 106a-106c to/from the word line drivers 202a-202c. Similarly, each bit line 108a-108b is coupled to a driver 206a-206b via selector gates 208a-208b. The current and/or resistance of a selected memory cell 100a-100f is measured by sensor amplifiers 210a-210b connected respectively to the bit lines 108a-108b. 
For simplicity, FIG. 4 illustrates a memory array having only two rows of memory cells 100 on two bit lines 108a-108b and three columns of memory cells 100 on three word lines 106a-106c. However, it should be understood that in practical applications, memory devices would have significantly more cells in an array. For example, an actual memory device may include several million cells 100 arranged in a number of subarrays.
While the overall operating scheme of the memory device 200 may be similar regardless of the type of variable resistance material used in the memory elements, much research of late has focused on memory devices using memory elements having doped chalcogenide materials as the variable resistance material. More specifically, memory cells having a variable resistance material formed of germanium-selenide glass having a stoichiometry of GexSe100-x, with x ranging from about 20 to about 43, have been shown to be particularly promising for providing a viable commercial alternative to traditional DRAMs.
Generally, a chalcogenide PCRAM cell having such stoichiometry has an initial and “off” state resistance of over 100 KΩ (e.g., 1 MΩ). To perform a write operation on a chalcogenide memory cell in its normal high resistive state, a voltage having at least a threshold potential is applied to the electrode serving as the anode, with the cathode held at the reference potential or ground. Upon applying the threshold level voltage, i.e., a write voltage, the resistance across the memory cell changes to a level dramatically reduced from the resistance in its normal state, to a resistance less than 100 KΩ (e.g., 20 KΩ), whereupon the cell is considered to be in the “on” state.
The PCRAM cell retains this new lower level of resistivity until the resistivity is changed by another qualifying voltage level applied to one of the electrodes of the cell. For example, the memory cell is returned to the high resistance state by applying an erase voltage thereto in the negative direction of the voltage applied in the write operation (to achieve the lower resistance state). The erase voltage may or may not be the same magnitude as the write voltage, but is at least of the same order of magnitude.
Although it is not clearly understood what change or changes are induced in the doped chalcogenide material by the application of the threshold potential to result in the stable low resistant state, it is believed that the metal ions incorporated into the chalcogenide material somehow become aligned into a low resistance conductive configuration between the electrodes once the applied voltage reaches the threshold level. At least two theories exist as to the precise nature of the alignment.
In one theory, the metal ions within the chalcogenide material begin to plate on the cathode and progress through the chalcogenide material toward the anode upon the application of the threshold voltage level of a write operation. The metal ions continue to agglomerate until a conductive dendrite or filament is extended between the electrodes to thereby interconnect the top and bottom electrodes to create an electrical short circuit. Upon application of the negative threshold voltage, the dendrite recedes from the anode as the metal ions return to solution in the chalcogenide material or return to the source layer.
In a second theory, when an initial write signal having a threshold positive voltage level is applied to the memory cell, at least one conductive channel is formed in the chalcogenide material, and the metal ions are caused to agglomerate along the channels(s), thereby lowering the electrical resistance across the cell. Upon application of an erase signal having a negative threshold voltage level to the cell, the conductive channel(s) remain in place, but the metal ions are caused to move away from the conductive channel(s), thereby raising the electrical resistance across the cell. Subsequent write signals cause the metal ions to re-agglomerate along the channel(s). The resistance through the cell in the write and erase states is thus determined by the amount of metal ions agglomerated along the channel(s).
When written to the low resistance state, chalcogenide PCRAM cells can retain this state for several hours, days, even weeks. In this regard, such PCRAM devices are relatively non-volatile as compared to DRAM devices. However, while chalcogenide PCRAM cells in the high resistance state are completely non-volatile, PCRAM cells written to the low resistance state will gradually lose their conductivity across the chalcogenide glass layer and drift towards the high resistance state after an extended period of time. In particular, it has been found that such chalcogenide PCRAM devices which are written according to high speed algorithms, i.e., using write voltages each having a pulse width of less than 100 ns, have a tendency to gradually lose their low resistance characteristic over time, e.g., a week. It is desirable to use such high speed algorithms to write PCRAM devices to accommodate the demands of current operating speeds in state-of-the-art processors. Accordingly, PCRAM memory devices operated using high speed algorithms should be intermittently refreshed to maintain optimal operation of the PCRAM memory devices.
Refresh operations are well known in DRAM devices. Specifically, in DRAM devices memory cells must be refreshed frequently, e.g., every 100 milliseconds, and regardless of the value stored in each cell. Additionally, DRAM cells are refreshed with every read operation. PCRAM cells, on the other hand, only require refreshing on a much more sporadic basis, e.g., once every several days to once per week. Moreover, only PCRAM cells written to the low resistance state need periodic refreshing, since PCRAM cells in the high resistance state are not at risk of losing their stored values. One feature resulting from these characteristics of PCRAM devices is that PCRAM devices do not require the dedicated circuitry necessary in DRAM devices to perform the refresh operations.
U.S. application Ser. No. 09/941,648, co-assigned to the assignee of the present invention, and the entire disclosure of which is hereby incorporated by reference, discloses a system and method for refreshing PCRAM memory cells by applying a positive voltage in the form of a sweep voltage, a pulse voltage, or a step voltage to the cells. The refresh voltages are applied at a magnitude less than that capable of writing a cell to the low resistance state, so that cells in the high resistance state are not affected by the refresh voltages.